Mapping Who Really Runs the Internet: Introducing “Borges”

Many measurements, policy debates, and incident reports treat networks (known as Autonomous Systems, or ASes) as independent actors. In practice, however, multiple ASes often belong to the same corporate group, shaping how we understand competition, resilience, and responsibility. Capturing these relationships reveals hidden interdependencies—from shared backhaul and submarine-cable capacity to joint procurement.

For example, Edgio—formed through the merger of Limelight and Edgecast—entered bankruptcy and sold off its assets, ultimately shutting down its content delivery network (CDN) in January 2025. Within three years, a once-prominent operator vanished, illustrating how quickly corporate relationships can shift. 

A new study from Northwestern University and collaborators offers a clearer view of who actually operates the Internet—associating the Internet’s underlying ASes with their real-world controlling parent organizations. 

The team’s system, Borges, combines WHOIS registry data, PeeringDB records, website signals, and large language models (LLMs) to uncover related networks that traditional methods often miss. Borges runs continuously at low cost—about US $3 per one-click run—providing an inexpensive way to keep the map current.

How Borges Works

Borges (Better ORGanizations Entities mappingS) improves AS→Organization mapping using three complementary clues:

• WHOIS & PeeringDB identifiers that link ASes to registered organizations.
  
• LLM-based extraction that finds sibling AS numbers (ASNs) and related networks from unstructured text (e.g., “notes” or “aka” fields).
  
• Website analysis that connects brands and subsidiaries via redirects, domain similarity, and favicons.
  

Borges reports high accuracy across its learned stages (0.947 for extracting sibling ASNs, 0.986 for detecting shared domains or favicons). On the new Organization Factor (θ) metric, a metric that quantifies how many more scattered independent networks are accurately grouped under a single parent company, Borges scores 0.3576, outperforming prior systems such as AS2Org (by 7%) and as2org+ (by 3.3%), yielding more realistic clustering of networks under their parent companies. Re-grouping previously scattered networks increases the recognized user base by roughly 192 million people (~5% of the global Internet) and expands national footprints for several providers.

Examples in Practice

• Sprint ↔ Cogent: Sprint’s metadata fields point to cogentco.com, reflecting Cogent’s acquisition of Sprint’s backbone.
  
• Limelight (AS22822) & Edgecast (AS15133) → Edgio: Both resolve to edg.io, capturing the merger and rebrand path.
  
• Clearwire → Sprint → T-Mobile: Historical redirects (clearwire.com → sprint.com → t-mobile.com) trace the full acquisition chain.
  
• Branding via favicons: Subsidiaries often reuse identical or near-identical favicons even when domains differ slightly, for example, Digicel and Claro, revealing shared corporate structure through visual branding.
  

Regular, low-cost reruns ensure Borges remains aligned with the Internet’s evolving corporate landscape.

Better Understanding = Better Decisions

Borges is designed to keep pace with that churn: lightweight to run, easy to schedule, and inexpensive to repeat, so the AS-to-organization map stays current as the Internet’s ownership landscape evolves. This higher fidelity of data allows us to better understand:

• Resilience and concentration. Consolidated ownership can be a source of strength or create single points of failure. Better maps sharpen risk assessments and diversification strategies, especially as corporate control shifts.
  
• Incident response. During outages, hijacks, or interference, knowing the actual operator family speeds root-cause analysis and accountability, provided mappings are refreshed as organizations change.
  
• Policy and transparency. Policymakers and advocates gain a clearer view of market power, mergers, and cross-border dependencies that shape user experience—updated as frequently as the landscape demands.

For more information, refer to our paper‘Learning AS-to-Organization Mappings with Borges,’ which we presented at IMC ’25. Our code and prompts are also open source, ensuring full reproducibility (one-click run, approximately US $3).

Esteban Carisimo is a Postdoctoral Researcher at AquaLab in the Department of Computer Science at Northwestern University.

The views expressed by the authors of this blog are their own and do not necessarily reflect the views of the Internet Society.

---

Image by Philip Neumann from Pixabay